The present invention relates to an arrangement and a method for decoding a digital signal, referred to hereinafter as data word.
Arrangements and methods of this type are used in message transmission and data storage for the reconstruction of a message transmitted as an analog or digital signal, referred to hereinafter as code word.
A code word is a data sequence which is generated by adding redundancy to information according to a given specification, referred to hereinafter as code.
An information source such as, for instance, a storage element or a transmitter transmits a code word to a receiver by means of a channel. If said channel is noisy, the code word can be damaged. The data word and the code word are then different.
The decoder generates an output signal with the aid of the data word (FIG. 1). Said output signal can comprise the code word, the information or parts of the code word and/or of the information.
The decoding method according to the invention is an error-correcting measure for determining the code word or the information or parts thereof with the aid of the data word. This makes use of the fact that the information is stored in the code word in a redundant manner.
A code word and a data word consist in each case of N symbols. The information consists of K symbols. In the case of a systematic coding, the code word consists of the information and a redundancy data sequence, such that the information is directly readable from the code word. In the case of a non-systematic coding, the information is not directly readable from the code word.
The code additionally defines the so-called minimum distance D. If the number of symbols damaged during transmission is a maximum of (D-1)/2 a suitable decoder can unambiguously reconstruct the code word or the information or parts thereof.
Reed-Muller codes are a family of linear error-correcting codes. Provided that the decoder used supports this, the structure of the Reed-Muller codes allows both the entire information and parts of the information to be decoded very rapidly. Moreover, Reed-Muller codes permit a broad use for different applications since the block length N and the number as to how many errors can be corrected can be tailored to the respective application.
This class of codes was developed in 1954 by Irving S. Reed and David E. Muller (1), (2) (all citations to references ( ) are with respect to the “Literature/References” section of the application). The binary Reed-Muller code was used by NASA in the Mariner Expeditions (1969 to 1976) to Mars in order to send the photographs taken of Mars to Earth.
Reed-Muller codes have been used to a lesser extent in recent times because their information rates are generally lower than in other classes of codes. Since their structure allows a parallel decoding of the information, this class of codes is increasingly of interest for multicore applications.
Different types of decoding can occasionally also be applied in combination to this code class: e.g. hard decision decoders based on majority logic in (2), (6), (7) or recursive soft decision decoders, e.g. in (3), (4).
Majority Logic decoding comprises a plurality of stages of majority decisions. In a majority decision, the value which occurs the most frequently is communicated. This type of decoding has the advantage that it can be implemented very simply and primarily at the hardware level. In addition, majority decoding is very fast. In contrast to decodings that utilize the occasionally cyclic or recursively established structure of a code, parallel decoding of the entire information is possible in the case of majority decoding, with the result that the decoding time can be shortened.
The Reed algorithm (2) is a majority logic hard decision decoding method for Reed-Muller codes. This algorithm consists of (R+1) levels of majority decisions, wherein R increases with increasing information length K and with block length N remaining the same. Reed's method therefore requires a multiplicity of decision modules and cannot decode in constant parallel time.
The method described by Dumer (3) involves recursive decoding by carrying out computation steps that are few in number but build sequentially on one another. This results in a lengthened decoding time since calculations cannot be performed in parallel. Moreover, this method can only be realized with relatively complex components or a multiplicity of components at the hardware level, since it operates with fixed-point numbers.
The patent specification WO 0176077 A2 presents a decoding scheme that focuses on serial decoding using latches. The individual bits are decoded with the aid of switching networks that occasionally contain majority gates. Concrete computation steps for the decoding are not mentioned; instead, reference is made to established methods described in (5).
Lin and Costello describe in detail in (5) the prior art with regard to decoders for Reed-Muller codes. All the decoders mentioned therein require significantly more majority gates for decoding. In particular Lin and Costello also mention the method from Chen (6), (7), which requires the fewest majority gates among the parallel decoders presented in (5).
The method (6), (7) described by Chen can be implemented in hardware, although with more components and an increased space requirement. In particular, more majority decisions have to be taken.
The patent specification U.S. Pat. No. 5,926,488 A claims decoding only for second-order Reed-Muller codes, for which R=2 holds true. The method presented therein is likewise a recursive method that requires a multiplicity of components for realization.
The patent U.S. Pat. No. 7,331,012 B2 claims a system for decoding Reed-Muller-coded signals. The decoder essentially consists of a soft decision algorithm based on the Reed algorithm.
Therefore, the object of the present invention is to provide an arrangement and a method for efficient parallel real-time decoding (error detection and correction) of a data word which can be realized in hardware and with a simpler architecture and fewer components in comparison with the prior art.
This object is achieved by means of an arrangement and method according to embodiments of the invention.
The linear code used in the present invention is a binary Reed-Muller code or a code equivalent thereto. N shall denote the length, D the distance and K the dimension of the code. Furthermore, E and F shall be defined as follows: E=D-2 and F=N/D. The natural number G shall describe how many of the N symbols of the data word are intended to be corrected. H shall describe a natural number between 1 and ED.
The arrangement according to the invention consists of the following components (FIG. 2):                a. N input terminals, I1 to IN,        b. a first level of ED summing modules, SI_11 to SI_ED, wherein each summing module of the first level is linked with F different input terminals and each input terminal is linked with E summing modules of the first level,        c. a first level of E decision modules MI_1 to MI_E, wherein each of the D inputs of each decision module is linked with a respective output from D different summing modules from the first level of the summing modules,        d. a second level of H summing modules, SII_1 to SII_H, wherein each summing module is linked with a decision module from the first level of the decision modules and a summing module from the first level of the summing modules,        e. a second level of G decision modules, MII_1 to MII_G, wherein each of the E inputs of each decision module is linked with a respective output from E different summing modules from the second level of the summing modules,        f. a third level of G summing modules, SIII_1 to SIII_G, wherein each summing module is linked with exactly one decision module from the second level of the decision modules and an input terminal,        g. G output terminals, O1 to OG, wherein each output terminal is linked with exactly one summing module from the third level of the summing modules.        h. The concrete interlinking NI_1 to NI_E of the N input terminals I1 to IN with the summing modules of the first level SI_11 to SI_ED and the concrete interlinking NII_1 to NII_E of the summing modules of the second level SII_11 to SII_ED with the decision modules of the second level MII_1 to MII_G depend on the concrete code.        
By way of example, the arrangement of the switching networks NI_1 to NI_E and NII_1 to NII_E for the Reed-Muller code RM(2,4) is illustrated and described in (FIG. 3). With the aid of the explanatory description on the basis of the exemplary embodiment RM(2,4), it is clearly evident to the person skilled in the art how the arrangements should be designed for other codes.
Preferably, G=N or G=K, such that either the entire code word or exclusively the entire information can be restored. Particularly preferably, H is equal to E·D.
The addition operations to be performed in the summing modules correspond to XOR operations. The summing modules of the first level 2) can be realized at the hardware level by odd-parity gates. The summing modules of the second level (4) and of the third level (6) can be implemented by XOR gates in the case of a binary code. Majority gates can be used for the decision modules of the first level (3) and second level (5). If E is equal to 2, then an AND gate can be used for the decision module.
Furthermore, the present invention relates to the method for decoding a data word with the aid of an error-correcting linear code. If the number of transferred symbols is disregarded, there are two types of modules which can be implemented in parallel or sequentially: I) summing modules add up in a binary manner all the symbols transferred to them and return the sum, that is to say one or zero; II) decision modules check for the frequency of the symbols transferred to it and return that symbol which occurs the most frequently.
The method according to the invention comprises the following steps, wherein the parameters N, E, D, F and G are as defined above:                a. reading in the data word, consisting of N symbols;        b. processing F symbols of the data word with the aid of a summing module to form a sum, wherein this step is performed E*D times by the same summing module or by different summing modules;        c. determining—on the basis of D sums from (b)—that sum which occurs the most frequently, wherein this step is performed E times by the same decision module or by different decision modules;        d. adding each sum determined in c) which is not equal to zero to up to D of the transferred sums from (b);        e. determining—on the basis of E sums calculated in (b) and/or (d)—that sum which occurs the most frequently, wherein this step is performed G times by the same decision module or by different decision modules;        f. adding each value determined in (e) which is not equal to zero to a symbol of the data word;        g. outputting the sums generated in (f).        
The arrangement according to the invention has the major advantage that the underlying architecture constitutes a switching network rather than a switching circuit, as known from the prior art, and can be constructed with a small number of simple components. By way of example, no register or storage elements, no flip-flops or clock generators are required. In comparison with the conventional decoding method of the Reed-Muller code, the invention uses only two levels of decision modules with a total of not more G+E decision modules. As a result, not only are components and space saved, but the decoding time is shortened as well, such that all errors can be corrected by parallel decoding in real time.
Further advantages, features and application possibilities of the invention are described below on the basis of the exemplary embodiment described below with reference to the figures.